Ethics Statement

This study analysed de-identified, publicly available mortality data from Brazil's national health information system. According to Brazilian National Health Council Resolution 510/2016 [29], research using exclusively publicly available secondary data without individual identification is exempt from ethical review by a Research Ethics Committee.

Study design and setting

We conducted a nationwide ecological time-series study in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [30]. The study examined associations between daily ambient temperature and respiratory mortality across Brazil from 1 January 2010–31 December 2020. We included 646 municipalities with estimated populations of at least 50,000 inhabitants, representing approximately 65% of the national population and spanning all five Brazilian macroregions. The geographic distribution of included municipalities and the participant flow diagram are shown in Fig 1. This population threshold was selected to ensure sufficient daily death counts for stable model estimation and to align with methodological standards in multi-city temperature-mortality studies [31,20].

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Study area and participant flow diagram.

(A) Geographic distribution of the 646 municipalities included in the analysis, colour-coded by Brazilian macroregion. Circle size is proportional to the total number of respiratory deaths during the study period (2010–2020). (B) Flow diagram showing municipality selection based on population threshold (≥50,000 inhabitants) and data completeness criteria. Base map shapefile obtained from the Brazilian Institute of Geography and Statistics (IBGE), available at https://geoftp.ibge.gov.br/organizacao_do_territorio/malhas_territoriais/malhas_municipais/municipio_2020/ (public domain).

https://doi.org/10.1371/journal.pclm.0000801.g001

Data sources

Mortality data were obtained from Brazil's Mortality Information System (Sistema de Informação sobre Mortalidade, SIM) via the DATASUS platform [32]. Daily counts of deaths with respiratory diseases as the underlying cause (International Classification of Diseases, 10th Revision, codes J00–J99) were extracted, along with decedent characteristics including age, sex, and municipality of residence. Daily meteorological data were obtained from the ERA5-Land reanalysis dataset [33], which provides gridded climate variables at approximately 9 km spatial resolution. For each municipality, we extracted daily mean temperature (°C), calculated as the average of 2-metre minimum and maximum temperatures, and relative humidity (%).

Climate data and exposure definition

The primary exposure was the daily mean ambient temperature. We selected mean temperature rather than minimum or maximum values based on three considerations: (1) mean temperature captures integrated daily thermal exposure more comprehensively than single extreme values; (2) mean temperature demonstrates stronger and more consistent associations with mortality in multi-country studies [5,31]; and (3) using a single metric facilitates comparison across municipalities with different diurnal temperature ranges. Furthermore, mean temperature captures the cumulative physiological burden of thermal exposure over a 24-hour period; sustained thermal stress—rather than brief temperature peaks—drives inflammatory responses and exacerbates chronic respiratory conditions [9,13]. In a sensitivity analysis, we also examined associations separately for minimum and maximum temperatures.

Outcome definition

The primary outcome was the daily count of deaths from respiratory diseases (ICD-10 J00–J99) occurring in each municipality during the study period. This broad category encompasses acute respiratory infections, influenza, pneumonia, chronic obstructive pulmonary disease, asthma, and other respiratory conditions.

Missing data

Mortality records with missing municipality codes (<0.1% of deaths) were excluded. Municipalities with more than 10% missing daily temperature data were excluded from analysis. Among included municipalities, the median proportion of days with missing climate data was 1.8% (interquartile range 0.5–3.2%). Missing climate values were not imputed; days with missing data were excluded from city-specific analyses.

Statistical analysis

First-stage analysis. For each municipality, we fitted a time-series regression model using the distributed lag non-linear model (DLNM) framework [34,35]. The model specification was:

where Yt is the daily death count, cb() is the cross-basis function for temperature with lag up to 21 days, ns() denotes natural cubic splines, DOW is the day of week, RH is the relative humidity, and Pop is the annual municipal population estimate from IBGE intercensal projections used as an offset to model mortality rates. The cross-basis function used natural cubic splines with 4 degrees of freedom for temperature and 4 for lag. Seasonal and long-term trends were controlled using natural cubic splines of time with 8 degrees of freedom per year (approximately one internal knot per 1.5 months), allowing flexible capture of seasonal patterns while avoiding overfitting. The quasi-Poisson family was used to account for overdispersion typically observed in daily mortality count data [36].

Second-stage analysis. Municipality-specific exposure-response associations, represented as regression coefficients from the cross-basis functions, were pooled using multivariate random-effects meta-analysis [37] to obtain overall and region-specific cumulative exposure-response curves. Between-municipality heterogeneity was quantified using the I² statistic, interpreted as minimal (<25%), moderate (25–50%), or substantial (>50%) [38,39].

Minimum mortality temperature and relative risks

The minimum mortality temperature (MMT), defined as the temperature at which mortality risk is lowest, was identified from each pooled exposure-response curve. Cumulative relative risks (RR) over lags 0–21 days were calculated at the 1st, 5th, 10th, 90th, 95th, and 99th percentiles of the national temperature distribution, with the MMT as reference. We selected multiple percentile thresholds based on methodological precedent in temperature-mortality research [5,31]: extreme percentiles (1st, 99th) capture rare but intense exposures, while moderate percentiles (5th, 10th, 90th, 95th) represent more common cold and heat exposures.

Attributable burden

Attributable fractions (AF) and attributable numbers (AN) of deaths due to non-optimal temperatures were calculated using the method described by Gasparrini and Leone [40]. The total temperature-attributable burden was partitioned into cold-attributable (temperatures below MMT) and heat-attributable (temperatures above MMT) components. Empirical 95% confidence intervals were estimated using Monte Carlo simulations (1,000 iterations).

Sensitivity analyses

We conducted four sensitivity analyses to assess robustness: (1) alternative percentile thresholds (5th/95th instead of 10th/90th) for defining temperature extremes; (2) alternative temperature metrics (minimum and maximum temperatures); (3) varying degrees of freedom for the temperature-lag cross-basis (3–5 df for each dimension); and (4) exclusion of the year 2020 to assess potential confounding by COVID-19-related mortality changes.

Software

All analyses were performed in R version 4.3.1 [41]. The dlnm package (version 2.4.7) was used for distributed lag non-linear modelling [34], mvmeta (version 1.0.3) for multivariate meta-analysis [37], and ggplot2 for data visualisation [42].

Patient and public involvement

This study analysed routinely collected administrative mortality data. Patients and members of the public were not directly involved in the design, conduct, or reporting of this research. However, the findings are intended to inform public health policies for climate adaptation that will benefit the general population.

Study population and mortality characteristics

The final analysis included 646 municipalities distributed across all five Brazilian macroregions during the period 2010–2020. These municipalities accounted for approximately 65% of the national population and spanned the full range of Brazilian climatic zones, from equatorial in the North to subtropical in the South.

A total of 1,087,094 respiratory deaths (ICD-10: J00–J99) were recorded during the study period. The regional distribution reflected both population density and data availability: the Southeast contributed 613,195 deaths (56.4%), followed by the Northeast with 211,007 (19.4%), the South with 136,576 (12.6%), the Center-West with 66,528 (6.1%), and the North with 59,788 (5.5%). The number of municipalities included per region ranged from 44 in the Center-West to 256 in the Southeast.

Temperature distribution

Daily mean temperature showed marked variation across the national territory, following expected latitudinal and altitudinal gradients. Nationally, temperatures ranged from 12.1°C at the 1st percentile to 30.3°C at the 99th percentile. The 10th and 90th percentiles, which defined moderate cold and heat exposure in our analysis, corresponded to 18.3°C and 28.4°C, respectively.

The North region experienced consistently warm temperatures throughout the year, while the South exhibited the widest seasonal amplitude and lowest winter temperatures. The Southeast and the Center-West showed intermediate patterns, and the Northeast displayed relatively stable warm conditions with lower variability than equatorial areas.

Overall temperature-mortality relationship

The pooled exposure-response analysis revealed a J-shaped relationship between daily mean temperature and respiratory mortality (Fig 2). The minimum mortality temperature (MMT), representing the thermal optimum, was estimated at 22.4°C, corresponding approximately to the 75th percentile of the national temperature distribution.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Exposure-response relationship between daily mean temperature and respiratory mortality.

(A) National pooled exposure-response curve showing the cumulative relative risk (RR) of respiratory mortality over lags 0–21 days as a function of daily mean temperature. The solid line represents the point estimate, and the shaded area indicates the 95% confidence interval. The vertical dashed line marks the minimum mortality temperature (MMT = 22.4°C). (B) Regional exposure-response curves for each Brazilian macroregion, demonstrating geographic heterogeneity in temperature-mortality associations. The J-shaped relationship is evident nationally, with steeper heat effects in tropical regions (North, Northeast) and more pronounced cold effects in subtropical regions (South).

https://doi.org/10.1371/journal.pclm.0000801.g002

Cold effects. Mortality risk increased progressively as temperatures decreased below the MMT. At the 1st percentile of temperature (12.1°C), the cumulative relative risk (RR) over lags 0–21 days was 1.29 (95% CI: 1.23–1.35), corresponding to a 29% increase in respiratory mortality relative to the MMT. The relationship showed a dose-response pattern: at the 5th percentile (16.3°C), RR was 1.10 (95% CI: 1.07–1.13), and at the 10th percentile (18.3°C), RR was 1.04 (95% CI: 1.03–1.06).

Heat effects. Heat-related mortality risk increased more steeply above the MMT than cold effects below it. At the 90th percentile (28.4°C), RR was 1.17 (95% CI: 1.13–1.21). Risk escalated substantially at higher temperatures: at the 95th percentile (29.0°C), RR reached 1.24 (95% CI: 1.20–1.28), and at the 99th percentile (30.3°C), RR was 1.43 (95% CI: 1.36–1.50), indicating a 43% increase in mortality during extreme heat events (Table 1).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. National temperature-mortality relationship: relative risks at temperature percentiles, Brazil, 2010–2020.

https://doi.org/10.1371/journal.pclm.0000801.t001

Model heterogeneity

The random-effects meta-analysis demonstrated low-to-moderate heterogeneity across municipalities (I² = 17.3%), indicating that while the overall exposure-response relationship was consistent nationally, some municipal-level variation existed. This relatively low heterogeneity supports the validity of pooled national estimates.

Regional variation

Regional meta-analyses revealed critical geographic differences in temperature-mortality relationships, reflecting climatic adaptation and population vulnerability patterns.

Minimum mortality temperature. The MMT varied substantially across regions (Fig 3A). The relationship between local climate (MMT) and temperature vulnerability showed distinct regional patterns (Fig 3B–C). Regional exposure-response curves demonstrated marked differences in curve shape across macroregions (Fig 2B). The Southeast (22.0°C) and South (22.9°C) exhibited MMT values close to the national estimate, consistent with their temperate and subtropical climates, where populations experience both cold and heat extremes. The North region showed a markedly higher MMT (32.2°C), suggesting physiological and behavioral adaptation to consistently warm conditions. The Northeast displayed an intermediate pattern (28.5°C). The Center-West region produced an MMT estimate at the lower boundary of the temperature distribution (0.5°C), a known limitation when the true optimum lies outside the observed range. This instability likely reflects two factors: the region's relatively narrow thermal variability compared to other macroregions, and the lower density of included municipalities (n = 44 versus n = 256 in the Southeast), which limits statistical power to identify a precise inflection point. Results for this region should therefore be interpreted with caution, and the AF estimates may be less reliable.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Regional heterogeneity in minimum mortality temperature and temperature vulnerability.

(A) Distribution of municipality-specific minimum mortality temperatures (MMT) by macroregion, showing the median (horizontal line), interquartile range (box), and range (whiskers). (B) Relationship between regional MMT and cold-attributable fraction, illustrating that regions with higher MMT (warmer climates) exhibit minimal cold-related mortality. (C) Relationship between regional MMT and heat-attributable fraction, demonstrating that equatorial regions with high MMT experience the greatest heat-related mortality burden.

https://doi.org/10.1371/journal.pclm.0000801.g003

Exposure-response curve shapes. The shape of regional exposure-response curves differed markedly (Fig 2). In the North and Northeast, curves were relatively flat at low temperatures but steepened at high temperatures, indicating predominant heat vulnerability. The South exhibited a more symmetric U-shaped curve, with substantial increases in mortality at both temperature extremes, though with a more significant cold component. The Southeast showed intermediate characteristics with both cold and heat effects evident.

Temperature-attributable mortality burden

Using the region-specific MMT as the reference, we estimated that 6.08% of all respiratory deaths during the study period were attributable to suboptimal temperatures, corresponding to 66,079 excess deaths over the 11 years (Table 2).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Temperature-attributable respiratory mortality by region, Brazil, 2010–2020.

https://doi.org/10.1371/journal.pclm.0000801.t002

National burden by temperature component. Heat exposure accounted for a substantially larger proportion of attributable mortality than cold at the national level: 4.27% (46,429 deaths) versus 1.81% (19,650 deaths). This finding reflects Brazil's predominantly tropical climate, where most of the population is regularly exposed to temperatures above their thermal optimum.

Regional patterns. The distribution of temperature-attributable mortality varied dramatically across regions, revealing distinct vulnerability profiles (Table 2):

Heat-dominant regions. The North exhibited the highest total attributable fraction (12.5%), with virtually all temperature-related mortality due to heat (AF heat: 12.5%; AF cold: < 0.01%; 7,477 excess deaths). The Northeast showed a similar pattern (AF total: 8.6%; AF heat: 8.6%; AF cold: 0.02%; 18,095 excess deaths from heat).

Cold-dominant region. In contrast, the South exhibited a reversed pattern, with cold-attributable mortality (5.97%; 8,159 deaths) substantially exceeding heat-attributable mortality (1.54%; 2,099 deaths), yielding a total AF of 7.5%. This reflects the region's subtropical climate, with significant exposure to winter cold.

Mixed patterns. The Southeast showed a more balanced distribution of cold- and heat-related effects (AF cold: 1.82%; AF heat: 2.57%; AF total: 4.39%), consistent with its diverse climate spanning tropical to subtropical zones. The Center-West showed predominantly heat effects (AF heat: 4.48%), though results should be interpreted with the methodological caution noted above.

Sensitivity analyses

The main findings were robust to alternative model specifications. Using the 5th and 95th percentiles as alternative thresholds for defining cold and heat exposure yielded consistent relative risk estimates. The J-shaped national exposure-response curve and the regional patterns of heat-dominant vulnerability in the North and Northeast versus cold-dominant vulnerability in the South remained stable across specifications. Additional sensitivity analyses examining different degrees of freedom for temporal control and alternative lag structures produced qualitatively similar results.

Summary of principal findings

This nationwide study of 646 Brazilian municipalities provides the first comprehensive assessment of temperature-attributable respiratory mortality across a tropical middle-income country. We observed a J-shaped exposure-response relationship with minimum mortality at 22.4°C, consistent with adaptation to local climate. Non-optimal temperatures accounted for 6.08% of respiratory deaths during 2010–2020, corresponding to 66,079 excess deaths over the study period—approximately 6,000 annually. Heat exposure accounted for a larger share of the attributable burden (4.27%; 46,429 deaths) than cold exposure (1.81%; 19,650 deaths), reflecting Brazil's predominantly tropical climate.

Regional heterogeneity was striking: the North exhibited almost exclusively heat-related mortality (attributable fraction 12.51%), whereas the South showed the opposite pattern, dominated by cold effects (5.97% cold versus 1.54% heat). The low heterogeneity observed nationally (I² = 17.3%) supports the validity of pooled estimates, while regional stratification revealed clinically meaningful geographic variation. These findings have direct implications for climate adaptation policy in tropical settings facing increasing temperature extremes.

Comparison with previous evidence

Our national attributable fraction of 6.08% falls within the range reported by multi-country collaborative studies. Gasparrini and colleagues, analysing 74 million deaths across 13 countries, estimated a global temperature-attributable mortality fraction of 7.71%, with most burden attributable to cold in temperate regions [5]. Zhao and colleagues reported that 9.43% of global deaths were attributable to non-optimal temperatures during 2000–2019, with substantial geographic heterogeneity [2]. A recent Multi-Country Multi-City (MCC) analysis of Central and South American countries found that tropical locations exhibited lower cold-attributable mortality (1.71%) than arid and temperate zones (5.10% and 5.29%, respectively). In comparison, heat effects were slightly higher in tropical climates (0.92%) [43].

The predominance of heat-attributable mortality in our study contrasts sharply with findings in temperate regions, where cold typically accounts for the larger share [5,11]. This pattern is biologically plausible given Brazil's climate: populations are more frequently exposed to temperatures above their thermal optimum, and the steep rise in mortality risk we observed at higher percentiles (RR = 1.43 at the 99th percentile versus RR = 1.29 at the 1st percentile) reflects limited physiological and infrastructural adaptation to heat extremes. Similar heat-dominant patterns have been documented in Thailand (attributable fraction 3.37%) and other equatorial settings [2,43].

Our minimum mortality temperature of 22.4°C—approximately at the 75th percentile of the national temperature distribution—aligns with global evidence that MMT varies geographically and reflects long-term population adaptation to local climate [44,45]. Tobías and colleagues, analysing 658 communities across 43 countries, found that MMT increased from alpine zones (13.0°C) through continental (19.3°C) and temperate (21.7°C) to tropical zones (26.5°C) [44]. Our national MMT falls between temperate and tropical benchmarks, consistent with Brazil's climatic diversity, which spans subtropical southern states to equatorial northern regions.

Previous Brazilian research was limited to single cities. Son and colleagues found that both cold and heat were associated with mortality in São Paulo, with cold effects predominating—consistent with our findings for the subtropical South region [27]. Gouveia and colleagues documented socioeconomic modification of temperature-mortality relationships in São Paulo, with more potent effects among deprived populations [26]. Silveira and colleagues reported significant effects of heat waves on cardiovascular and respiratory mortality in Rio de Janeiro [46]. Our study extends this evidence nationally, demonstrating that the patterns observed in southeastern cities are not generalizable to tropical northern Brazil, where heat vulnerability dominates.

The cold paradox in warm regions

A noteworthy finding is the substantial cold-attributable mortality burden even in tropical regions. In the North, the 10th percentile of daily mean temperature corresponds to approximately 24°C—considerably higher than conventional cold thresholds in temperate climates (typically below 10°C) but representing the lower extreme of local thermal experience. This phenomenon, sometimes termed the “cold paradox,” reflects heightened vulnerability to temperatures that would be considered mild elsewhere [11,47].

Several mechanisms may explain this pattern. First, housing in tropical Brazil is typically designed for heat dissipation rather than cold protection. Dwellings feature open ventilation, lightweight construction materials, single-pane windows, and minimal thermal insulation, leaving occupants physiologically and structurally unprepared for temperature drops [48]. Heating systems are virtually absent in residential buildings throughout tropical and equatorial Brazil—a stark contrast to temperate regions where indoor climate control buffers outdoor temperature extremes. Second, populations acclimatized to consistently warm conditions may lack the physiological cold-acclimatization responses (such as enhanced peripheral vasoconstriction and non-shivering thermogenesis) that protect populations in colder climates [47]. Third, healthcare systems in northern Brazil may be less equipped to manage cold-related respiratory exacerbations, which are rare events in these settings.

The Eurowinter study demonstrated that populations in warmer European countries (Greece, southern Italy) exhibited greater vulnerability to moderate cold than those in colder northern countries, despite—or perhaps because of—their milder baseline climates [12]. Our findings extend this observation to tropical settings, suggesting that climate adaptation policies in Brazil should not neglect cold-weather preparedness even in predominantly warm regions.

Heat vulnerability and regional disparities

The North (12.51%) and Northeast (8.60%) exhibited the highest temperature-attributable mortality fractions, driven almost entirely by heat effects. Several factors may contribute to this regional vulnerability pattern.

First, these regions experience consistently high temperatures with limited seasonal variation, resulting in sustained thermal stress rather than discrete heat events. The narrow temperature range reduces opportunities for physiological acclimatization to temperature extremes [49]. Second, socioeconomic factors compound physiological vulnerability: the North and Northeast are Brazil's poorest regions, with higher rates of informal housing, limited access to air conditioning, and constrained healthcare infrastructure [48]. Urban heat island effects further amplify exposure in rapidly growing cities such as Manaus and Fortaleza, where impervious surfaces and reduced vegetation elevate local temperatures [14].

Third, the demographic composition of these regions may increase susceptibility. While younger on average than populations in southern Brazil, northern populations include substantial proportions of outdoor workers in agriculture, construction, and the informal sector who face occupational heat exposure and limited capacity for behavioural adaptation (e.g., seeking shade, reducing activity) [50]. The near-absence of cold-attributable mortality in the North (AF < 0.01%) confirms that heat, not cold, is the dominant temperature-related health threat in equatorial Brazil.

Beyond structural deficiencies, indoor environmental quality in low-income housing—including inadequate ventilation, indoor dampness, and mould growth—may further compromise respiratory health and amplify vulnerability to thermal extremes. Additionally, limited access to primary healthcare in rural and peri-urban areas of northern Brazil may delay recognition and treatment of temperature-related respiratory exacerbations. These housing and healthcare pathways warrant investigation through individual-level studies capable of disentangling structural vulnerability from physiological susceptibility.

The Southeast region, which accounted for the largest absolute number of attributable deaths (26,920), showed a more balanced distribution between cold (1.82%) and heat (2.57%) effects. This pattern reflects the region's climatic diversity—spanning tropical coastal areas to subtropical highlands—and its large elderly population concentrated in major metropolitan areas. São Paulo and Rio de Janeiro, Brazil's largest cities, experience significant temperature variability and have been the focus of previous single-city studies documenting both cold and heat effects [26,27,46].

Biological mechanisms

The biological plausibility of our findings is supported by established pathophysiological mechanisms linking temperature extremes to respiratory mortality.

Cold exposure impairs respiratory defence mechanisms at multiple levels. Breathing cold air triggers bronchoconstriction and increases airway resistance, particularly in individuals with asthma or chronic obstructive pulmonary disease [9]. Cold temperatures damage the respiratory epithelium, impair mucociliary clearance, and increase mucus viscosity, creating favourable conditions for bacterial colonization and secondary infection [7]. Additionally, cold and low humidity enhance the stability and transmissibility of enveloped respiratory viruses—including influenza, respiratory syncytial virus, and coronaviruses—contributing to winter epidemics in temperate regions [8]. The prolonged lag structure we observed for cold effects (extending to 21 days) is consistent with these mechanisms, which involve delayed inflammatory cascades, opportunistic infections, and exacerbations of underlying chronic disease.

Heat exposure affects respiratory health through distinct but equally essential pathways. Acute heat stress causes hyperventilation, airway desiccation, and impaired gas exchange, exacerbating respiratory insufficiency in vulnerable individuals [13]. High temperatures increase ground-level ozone formation through photochemical reactions, with well-documented adverse effects on lung function and respiratory mortality [15,16]. Heat waves may also trigger systemic inflammatory responses that compound respiratory compromise [51]. The rapid onset of heat effects we observed (peaking within 0–3 days) reflects these acute physiological responses, in contrast to the more protracted time course of cold-related mortality.

Strengths and limitations

This study has several methodological strengths. First, the nationwide scope—encompassing 646 municipalities and more than 1 million respiratory deaths over 11 years—provides robust statistical power and broad geographic representation, rarely achieved in temperature-mortality studies in low- and middle-income countries. Second, the two-stage analytical framework, which uses distributed lag non-linear models and multivariate meta-analysis, represents current best practice, allowing flexible characterisation of non-linear and delayed exposure-response associations while appropriately accounting for heterogeneity [5,37]. Third, the relatively low statistical heterogeneity (I² = 17.3%) supports the validity of pooled national estimates, while regional stratification captures clinically meaningful geographic variation. Fourth, we employed pooled percentile-based knots across all municipalities, ensuring comparability of exposure-response functions and avoiding overfitting to local temperature distributions.

Several limitations warrant consideration. First, the ecological design precludes individual-level causal inference, and unmeasured confounders—including air pollution co-exposures, influenza circulation, and individual-level socioeconomic factors—may have influenced our estimates. Although we adjusted for temporal trends and seasonality, residual confounding cannot be excluded. Second, our analysis was restricted to municipalities with populations of at least 50,000, potentially limiting generalisability to rural areas, where healthcare access and completeness of vital registration may differ. Third, the Centre-West region produced minimum-mortality temperature estimates at the boundary of the fitted curve (within 0.5°C of the exposure range limit), indicating model instability; results for this region should be interpreted with caution.

Fourth, we used ambient temperature data from municipal weather stations, which may not fully capture individual-level thermal exposure—particularly in urban areas where microclimate variation and indoor environments modify actual exposure. Fifth, the use of daily mean temperature may obscure biologically relevant intraday thermal variation. Days with identical mean temperatures may reflect either stable conditions or sharp diurnal fluctuations; physiological responses to persistent cold differ from those to sudden thermal shock. However, sensitivity analyses using minimum and maximum temperatures yielded consistent results, suggesting robustness to the choice of temperature metric. Sixth, race and ethnicity classifications were based on death certificate records, which may be subject to misclassification, particularly for Indigenous populations.

Public health implications

Our findings support several evidence-based recommendations for climate adaptation policy in Brazil and similar tropical middle-income countries.

First, the substantial burden of temperature-attributable respiratory mortality—over 6,000 deaths annually—underscores the need for integrated climate-health surveillance systems. Brazil's existing respiratory disease monitoring infrastructure could be augmented with temperature-based early warning triggers, enabling proactive public health responses during thermal extremes [52].

Second, the striking regional heterogeneity in vulnerability profiles argues against adopting uniform national policies. Heat action plans are most urgently needed in northern Brazil, where attributable fractions exceed 12%, and populations lack physiological and infrastructural adaptation to extreme heat. Conversely, southern regions require continued investment in cold-weather preparedness despite their subtropical climate—our finding that cold accounts for nearly 6% of respiratory deaths in the South challenges assumptions that cold is not a health threat in Brazil.

Third, adaptation strategies should prioritise vulnerable populations. The elderly, who account for approximately 75% of respiratory deaths, face compounded vulnerability from age-related thermoregulatory decline and high prevalence of chronic respiratory disease [13,51]. Community-based interventions—including cooling centres during heat waves, home visits to isolated elderly during cold spells, and public awareness campaigns about thermal risks—have demonstrated effectiveness in reducing temperature-related mortality in other settings and merit evaluation in the Brazilian context [52].

Fourth, our findings inform Brazil's National Adaptation Plan to Climate Change by quantifying region-specific health risks [28]. As climate change increases the frequency and intensity of temperature extremes, particularly heat waves, the burden of temperature-attributable mortality is projected to grow substantially in tropical regions [4,6,31]. Investment in heat-resilient urban design, expansion of air-conditioning access in healthcare facilities and vulnerable households, and strengthening of the healthcare system's surge capacity will be essential components of effective adaptation.